WHERE AI HELPS PEOPLE AT WORK Recorded Syaxis example Created and revised: 31 October 2026 WHAT THIS EXAMPLE SHOWS A complete eight-slide presentation created in Syaxis from a short brief and the inputs below. Syaxis handled the research or source reading, writing and slide design. The operator answered its setup questions and used the same chat for the revisions recorded here. All slide creation and edits were made in Syaxis; the exported deck was not rebuilt or manually redrawn elsewhere. This is a selected, reviewed and revised product demonstration, not a first-pass accuracy benchmark. The final PDF, PowerPoint export and preview come from the same approved presentation. INPUTS No uploaded files or prescribed source links. Syaxis selected the research. SETUP ANSWERS Summary of the selected answers to Syaxis questions, not a verbatim chat transcript: 8 slides, broad overview; knowledge work/writing, customer service/sales, software/data/technical work, risks/implementation. ACTUAL SUBMITTED PROMPTS Initial brief Research where AI helps people at work and create a presentation for business leaders. An energetic science-magazine style with a couple of evidence-led visuals would work well. Revision 1 On slide 5, change “GitHub Copilot experiment, 2023” to “GitHub Copilot paper, 2023”. Keep everything else unchanged. OUTPUT AND REVIEW Eight slides. The final native slide images were visually reviewed, and the actual PowerPoint export was rendered in LibreOffice and checked. Source review used AI-assisted checks against the sources below. The example was selected for this showcase after visual review. PowerPoint includes editable text and supported shapes alongside raster visuals. These files do not contain native PowerPoint chart or table objects. Speaker notes contain slide numbers, not a bibliography or substantive presenter notes. Microsoft PowerPoint application editing was not tested. Use the PDF for the closest representation of the reviewed slide design, and this record for detailed source provenance. workplace-ai-research.pdf SHA-256: f7951cafe5715cb2947de2be1e46fce614cb62543b4754c74f582430b26f0688 workplace-ai-research.pptx SHA-256: 824d20e449369044475253257727a13c6c5d46757005a88a655cfc4f2320c97c SOURCES AND EVIDENCE Workplace AI research sources and evidence These are the sources for the eight-slide “Where AI helps people at work” briefing. Results concern specific tasks, workers and tools, not whole-job productivity or Syaxis performance. This briefing uses the published 2025 support study, not its earlier 2023 working-paper statistics. Writing Shakked Noy and Whitney Zhang, Experimental evidence on the productivity effects of generative artificial intelligence, Science, 14 July 2023. https://doi.org/10.1126/science.adh2586 An experiment with 453 college-educated professionals assigned occupation-specific writing tasks found 40% less time and 18% higher assessed quality with ChatGPT access. These are task outcomes, not a measure of every part of those jobs. The publisher page returned an access error during checking; the headline findings were corroborated through MIT's institutional account of its researchers' study: https://news.mit.edu/2023/study-finds-chatgpt-boosts-worker-productivity-writing-0714 Customer support Erik Brynjolfsson, Danielle Li and Lindsey Raymond, Generative AI at Work, The Quarterly Journal of Economics, volume 140, issue 2, May 2025, pages 889-942; published online 4 February 2025. https://academic.oup.com/qje/article/140/2/889/7990658 https://doi.org/10.1093/qje/qjae044 The published article reports 5,172 agents and an average 15% increase in issues resolved per hour. Gains vary and concentrate among less-experienced/lower-skilled workers. The study concerns one company's support setting; it does not establish sales revenue gains, aggregate employment effects or wage changes. The earlier working-paper values of 5,179 and 14% are not the version used here. Coding with Copilot Sida Peng, Eirini Kalliamvakou, Peter Cihon and Mert Demirer, The Impact of AI on Developer Productivity: Evidence from GitHub Copilot, arXiv version 1, 13 February 2023. https://arxiv.org/html/2302.06590v1 Ninety-five professional developers were recruited for a specified JavaScript HTTP-server task. The reported 55.8% faster completion is a completion-time estimate among those completing the task, not proof of equivalent gains across software engineering. The experiment ran from 15 May to 20 June 2022. “2023” is the paper year, not the experiment year. Experienced developers on their own repositories Joel Becker, Nate Rush, Beth Barnes and David Rein, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, METR, 10 July 2025. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/ The randomized study involved 16 experienced open-source developers and 246 real repository tasks. Allowing early-2025 AI tools increased completion time by 19% in this setting. It is not a finding about all developers or today's latest tools. The source now links later research; that later evidence is not the basis of this slide. Interpretation and recommendations The two coding results differ in task, population, tools and conditions. They are not a head-to-head product comparison. Do not combine the studies' percentages into a universal productivity or ROI number. Time reduction, rated quality and issues per hour are different measures. Checkable tasks, source verification, restricted permissions, human approval, training and a bounded baseline-versus-pilot comparison are recommendations synthesized for the briefing. They are not measured Syaxis outcomes, guaranteed controls or a headcount forecast. Any sales-enablement application is an opportunity to test, not proven revenue. Evaluate quality, full workflow time including review and debugging, costs and incidents before expanding. The academic studies are linked, not reproduced in full in this package.